Thanks. I pushed the fix to git branches. It will be included in future releases (but 1.11.3 isn't planned anytime soon).
It might be good to report a bug to VMware. I don't think they are supposed to advertise the x2APIC CPU feature unless they support CPUID 0xb leaf. Brice Le 03/02/2016 05:45, Jianjun Wen a écrit : > Confirmed! > This patch fixes the problem. > > Thanks a lot! > Jianjun > > On Tue, Feb 2, 2016 at 9:05 AM, Brice Goglin <brice.gog...@inria.fr > <mailto:brice.gog...@inria.fr>> wrote: > > Does this patch help? > > diff --git a/src/topology-x86.c b/src/topology-x86.c > index efd4300..a602121 100644 > --- a/src/topology-x86.c > +++ b/src/topology-x86.c > @@ -403,7 +403,7 @@ static void look_proc(struct hwloc_backend *backend, > struct procinfo *infos, uns > /* Get package/core/thread information from cpuid 0x0b > * (Intel x2APIC) > */ > - if (cpuid_type == intel && has_x2apic(features)) { > + if (cpuid_type == intel && highest_cpuid >= 0x0b && > has_x2apic(features)) { > unsigned level, apic_nextshift, apic_number, apic_type, apic_id = 0, > apic_shift = 0, id; > for (level = 0; ; level++) { > ecx = level; > > It looks like VMware reports that the virtual reports x2APIC > feature without 0xb CPUID. This looks buggy, but can be worked around. > > Brice > > > > > > Le 02/02/2016 05:50, Jianjun Wen a écrit : >> Hi Brice, >> Oh, didn't realize that. Only master has the gatther-cpuid. >> >> Attached. >> >> BTW, /proc/cpuinfo contain a field called flags. If it is an vm, >> hypervisor will be there. >> sudo dmidecode -s system-product-name >> will output >> VMware Virtual Platform >> >> Jianjun >> >> On Mon, Feb 1, 2016 at 12:26 AM, Brice Goglin >> <brice.gog...@inria.fr <mailto:brice.gog...@inria.fr>> wrote: >> >> Looks like you ran hwloc-gather-topology instead of >> hwloc-gather-cpuid? >> By the way, in (4) >> tar cfj cpuid.tbz2 foo >> should be >> tar cfj cpuid.tbz2 cpuid >> >> >> >> >> Le 01/02/2016 07:20, Jianjun Wen a écrit : >>> Hi Brice, >>> Thanks for the workaround -- it works very good. >>> >>> Attached please find the two output file after run >>> hwloc-gather-cpuid. >>> Let me after this is fixed! >>> >>> thanks, >>> Jianjun >>> >>> On Sun, Jan 31, 2016 at 9:48 PM, Brice Goglin >>> <brice.gog...@inria.fr <mailto:brice.gog...@inria.fr>> wrote: >>> >>> Thanks for the debugging. I guess VMware doesn't >>> properly emulate the CPUID instruction. >>> >>> Please do: >>> 1) take a tarball from git master at >>> https://ci.inria.fr/hwloc/job/master-0-tarball/ and build it >>> 2) export HWLOC_COMPONENTS=-x86 in your terminal >>> 3) do utils/hwloc/hwloc-gather-cpuid >>> 4) tar cfj cpuid.tbz2 foo and send that cpuid.tbz2 >>> >>> Step (3) might do an infinite loop for the same reason, >>> please replace >>> for(i=0; ; i++) { >>> with >>> for(i=0; i<10; i++) { >>> everywhere in utils/hwloc/hwloc-gather-cpuid.c >>> >>> This tarball will help me find what's buggy in VMware >>> CPUID instruction. >>> >>> >>> In the meantime, you can fix your hwloc by exporting >>> HWLOC_COMPONENTS=-x86 in your environment. >>> >>> If somebody knows how do detect vmware by looking under >>> /proc or /sys, we could use that to automatically set >>> that environment variable. >>> >>> thanks >>> Brice >>> >>> >>> >>> >>> >>> Le 01/02/2016 05:59, Jianjun Wen a écrit : >>>> I did a debug build. Found it loops forever in this >>>> loop in topology-x86.c:404. >>>> >>>> >>>> /* Get package/core/thread information from cpuid 0x0b >>>> * (Intel x2APIC) >>>> */ >>>> if (cpuid_type == intel && has_x2apic(features)) { >>>> unsigned level, apic_nextshift, apic_number, >>>> apic_type, apic_id = 0, apic_shift = 0, id; >>>> for (level = 0; ; level++) { >>>> ecx = level; >>>> eax = 0x0b; >>>> hwloc_x86_cpuid(&eax, &ebx, &ecx, &edx); >>>> if (!eax && !ebx) >>>> break; >>>> } >>>> >>>> On Sun, Jan 31, 2016 at 8:30 PM, Christopher Samuel >>>> <sam...@unimelb.edu.au <mailto:sam...@unimelb.edu.au>> >>>> wrote: >>>> >>>> On 01/02/16 15:09, Jianjun Wen wrote: >>>> >>>> > 0x00007ffff7bce13c in look_proc () from >>>> /lib64/libhwloc.so.5 >>>> > >>>> > Always the same place. >>>> >>>> pstack on the process when stuck might give more of >>>> an insight as it >>>> should give more of a stack trace. >>>> >>>> Also running lstopo under strace should show what >>>> it is trying to do at >>>> that point. >>>> >>>> All the best, >>>> Chris >>>> -- >>>> Christopher Samuel Senior Systems Administrator >>>> VLSCI - Victorian Life Sciences Computation Initiative >>>> Email: sam...@unimelb.edu.au >>>> <mailto:sam...@unimelb.edu.au> Phone: +61 (0)3 903 >>>> 55545 <tel:%2B61%20%280%293%20903%2055545> >>>> http://www.vlsci.org.au/ http://twitter.com/vlsci >>>> >>>> _______________________________________________ >>>> hwloc-users mailing list >>>> hwloc-us...@open-mpi.org >>>> <mailto:hwloc-us...@open-mpi.org> >>>> Subscription: >>>> http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-users >>>> Link to this post: >>>> >>>> http://www.open-mpi.org/community/lists/hwloc-users/2016/01/1251.php >>>> >>>> >>>> >>>> >>>> -- >>>> -Jianjun Wen >>>> Wancube.com - 3D photography >>>> Phone: 408 888 7023 <tel:408%20888%207023> >>>> >>>> >>>> _______________________________________________ >>>> hwloc-users mailing list hwloc-us...@open-mpi.org >>>> <mailto:hwloc-us...@open-mpi.org> Subscription: >>>> http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-users >>>> >>>> Link to this post: >>>> http://www.open-mpi.org/community/lists/hwloc-users/2016/01/1252.php >>> >>> >>> _______________________________________________ >>> hwloc-users mailing list >>> hwloc-us...@open-mpi.org <mailto:hwloc-us...@open-mpi.org> >>> Subscription: >>> http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-users >>> Link to this post: >>> >>> http://www.open-mpi.org/community/lists/hwloc-users/2016/02/1254.php >>> >>> >>> >>> >>> -- >>> -Jianjun Wen >>> Wancube.com - 3D photography >>> Phone: 408 888 7023 <tel:408%20888%207023> >>> >>> >>> _______________________________________________ >>> hwloc-users mailing list >>> hwloc-us...@open-mpi.org <mailto:hwloc-us...@open-mpi.org> >>> Subscription: >>> http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-users >>> Link to this post: >>> http://www.open-mpi.org/community/lists/hwloc-users/2016/02/1254.php >> >> >> >> >> -- >> -Jianjun Wen >> Wancube.com - 3D photography >> Phone: 408 888 7023 <tel:408%20888%207023> > > > > > -- > -Jianjun Wen > Wancube.com - 3D photography > Phone: 408 888 7023
