Re: [PATCH v7 07/13] hw/386: Add EPYC mode topology decoding functions

[Eduardo Habkost]

On Wed, Jun 03, 2020 at 10:38:46AM -0500, Babu Moger wrote:
> 
> 
> On 6/3/20 10:31 AM, Eduardo Habkost wrote:
> > On Wed, Jun 03, 2020 at 10:22:10AM -0500, Babu Moger wrote:
> >>
> >>
> >>> -----Original Message-----
> >>> From: Eduardo Habkost <ehabkost@redhat.com>
> >>> Sent: Tuesday, June 2, 2020 6:01 PM
> >>> To: Moger, Babu <Babu.Moger@amd.com>
> >>> Cc: marcel.apfelbaum@gmail.com; pbonzini@redhat.com; rth@twiddle.net;
> >>> mst@redhat.com; imammedo@redhat.com; qemu-devel@nongnu.org
> >>> Subject: Re: [PATCH v7 07/13] hw/386: Add EPYC mode topology decoding
> >>> functions
> >>>
> >>> On Tue, Jun 02, 2020 at 04:59:19PM -0500, Babu Moger wrote:
> >>>>
> >>>>
> >>>>> -----Original Message-----
> >>>>> From: Eduardo Habkost <ehabkost@redhat.com>
> >>>>> Sent: Tuesday, June 2, 2020 12:19 PM
> >>>>> To: Moger, Babu <Babu.Moger@amd.com>
> >>>>> Cc: marcel.apfelbaum@gmail.com; pbonzini@redhat.com; rth@twiddle.net;
> >>>>> mst@redhat.com; imammedo@redhat.com; qemu-devel@nongnu.org
> >>>>> Subject: Re: [PATCH v7 07/13] hw/386: Add EPYC mode topology decoding
> >>>>> functions
> >>>>>
> >>>>> Hi,
> >>>>>
> >>>>> It looks like this series breaks -device and CPU hotplug:
> >>>>>
> >>>>> On Wed, Mar 11, 2020 at 05:53:34PM -0500, Babu Moger wrote:
> >>>>>> These functions add support for building EPYC mode topology given the
> >>> smp
> >>>>>> details like numa nodes, cores, threads and sockets.
> >>>>>>
> >>>>>> The new apic id decoding is mostly similar to current apic id decoding
> >>>>>> except that it adds a new field node_id when numa configured. Removes
> >>> all
> >>>>>> the hardcoded values. Subsequent patches will use these functions to 
> >>>>>> build
> >>>>>> the topology.
> >>>>>>
> >>>>>> Following functions are added.
> >>>>>> apicid_llc_width_epyc
> >>>>>> apicid_llc_offset_epyc
> >>>>>> apicid_pkg_offset_epyc
> >>>>>> apicid_from_topo_ids_epyc
> >>>>>> x86_topo_ids_from_idx_epyc
> >>>>>> x86_topo_ids_from_apicid_epyc
> >>>>>> x86_apicid_from_cpu_idx_epyc
> >>>>>>
> >>>>>> The topology details are available in Processor Programming Reference
> >>> (PPR)
> >>>>>> for AMD Family 17h Model 01h, Revision B1 Processors. The revision
> >>> guides
> >>>>> are
> >>>>>> available from the bugzilla Link below.
> >>>>>> Link:
> >>>>>
> >>> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fbugzilla.
> >>>>>
> >>> kernel.org%2Fshow_bug.cgi%3Fid%3D206537&amp;data=02%7C01%7Cbabu.m
> >>>>>
> >>> oger%40amd.com%7C3487f40d37df4d59097d08d807190248%7C3dd8961fe488
> >>>>>
> >>> 4e608e11a82d994e183d%7C0%7C0%7C637267151289763739&amp;sdata=wE0
> >>>>> ukXIVh0l5eNQWsv6VDE9UZEVJmisofaW192gcZAI%3D&amp;reserved=0
> >>>>>>
> >>>>>> Signed-off-by: Babu Moger <babu.moger@amd.com>
> >>>>>> Acked-by: Igor Mammedov <imammedo@redhat.com>
> >>>>>> Acked-by: Michael S. Tsirkin <mst@redhat.com>
> >>>>>> ---
> >>>>> [...]
> >>>>>>  typedef struct X86CPUTopoIDs {
> >>>>>>      unsigned pkg_id;
> >>>>>> +    unsigned node_id;
> >>>>>
> >>>>> You have added a new field here.
> >>>>>
> >>>>>>      unsigned die_id;
> >>>>>>      unsigned core_id;
> >>>>>>      unsigned smt_id;
> >>>>> [...]
> >>>>>> +static inline apic_id_t
> >>>>>> +x86_apicid_from_topo_ids_epyc(X86CPUTopoInfo *topo_info,
> >>>>>> +                              const X86CPUTopoIDs *topo_ids)
> >>>>>> +{
> >>>>>> +    return (topo_ids->pkg_id  << apicid_pkg_offset_epyc(topo_info)) |
> >>>>>> +           (topo_ids->node_id << apicid_node_offset_epyc(topo_info)) |
> >>>>>
> >>>>> You are using the new field here.
> >>>>>
> >>>>>> +           (topo_ids->die_id  << apicid_die_offset(topo_info)) |
> >>>>>> +           (topo_ids->core_id << apicid_core_offset(topo_info)) |
> >>>>>> +           topo_ids->smt_id;
> >>>>>> +}
> >>>>>
> >>>>> But you are not initializing node_id in one caller of 
> >>>>> apicid_from_topo_ids():
> >>>>>
> >>>>> static void pc_cpu_pre_plug(HotplugHandler *hotplug_dev,
> >>>>>                             DeviceState *dev, Error **errp)
> >>>>> {
> >>>>>     [...]
> >>>>>     X86CPUTopoIDs topo_ids;
> >>>>>     [...]
> >>>>>     if (cpu->apic_id == UNASSIGNED_APIC_ID) {
> >>>>>         [...]
> >>>>>         topo_ids.pkg_id = cpu->socket_id;
> >>>>>         topo_ids.die_id = cpu->die_id;
> >>>>>         topo_ids.core_id = cpu->core_id;
> >>>>>         topo_ids.smt_id = cpu->thread_id;
> >>>>>         cpu->apic_id = x86ms->apicid_from_topo_ids(&topo_info, 
> >>>>> &topo_ids);
> >>>>>     }
> >>>>>     [...]
> >>>>> }
> >>>>>
> >>>>> Result: -device is broken when using -cpu EPYC:
> >>>>>
> >>>>>   $ qemu-system-x86_64 -machine q35,accel=kvm -smp
> >>>>> 1,maxcpus=2,cores=1,threads=1,sockets=2 -cpu EPYC -device EPYC-x86_64-
> >>>>> cpu,core-id=0,socket-id=1,thread-id=0
> >>>
> >>> [1]
> >>>
> >>>>>   qemu-system-x86_64: -device EPYC-x86_64-cpu,core-id=0,socket-
> >>> id=1,thread-
> >>>>> id=0: Invalid CPU [socket: 21855, die: 0, core: 0, thread: 0] with APIC 
> >>>>> ID
> >>> 21855,
> >>>>> valid index range 0:1
> >>>>>
> >>>>> This happens because APIC ID is calculated using uninitialized
> >>>>> memory.
> >>>> This patch should initialize the node_id. But I am not sure how to
> >>>> reproduce the bug. Can you please send me the full command line to
> >>>> reproduce the problem. Also test different options.
> >>>
> >>> The full command line is above[1].
> >>
> >> I just picked up the latest tree from
> >> git clone 
> >> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgit.qemu.org%2Fgit%2Fqemu.git&amp;data=02%7C01%7Cbabu.moger%40amd.com%7C7752c36b7ad24df4039d08d807d334fa%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637267951007572953&amp;sdata=69rG4w79FnLs9bhNmroxsRckzmy%2BGuH1dMjyP8PPRUo%3D&amp;reserved=0
> >>  qemu
> >> Not seeing the problem.
> >>
> >> ./x86_64-softmmu/qemu-system-x86_64 -machine q35,accel=kvm -smp
> >> 1,maxcpus=2,cores=1,threads=1,sockets=2 -cpu EPYC -device
> >> EPYC-x86_64-cpu,core-id=0,socket-id=1,thread-id=0
> >> VNC server running on ::1:5900
> >>
> >> It appears to run ok.
> 
> [2]
> 
> >>
> >>>
> >>>
> >>>>
> >>>> diff --git a/hw/i386/pc.c b/hw/i386/pc.c
> >>>> index 2128f3d6fe..047b4b9391 100644
> >>>> --- a/hw/i386/pc.c
> >>>> +++ b/hw/i386/pc.c
> >>>> @@ -1539,6 +1539,9 @@ static void pc_cpu_pre_plug(HotplugHandler
> >>> *hotplug_dev,
> >>>>      if (cpu->apic_id == UNASSIGNED_APIC_ID) {
> >>>>          int max_socket = (ms->smp.max_cpus - 1) /
> >>>>                                  smp_threads / smp_cores / 
> >>>> x86ms->smp_dies;
> >>>
> >>> So, here's the input you are using to calculate topo_ids.node_id:
> >>>
> >>>> +        unsigned nr_nodes = MAX(topo_info.nodes_per_pkg, 1);
> >>>
> >>> When is topo_info.nodes_per_pkg allowed to be 0?
> >>
> >> It will be zero if numa is not configured. Node_id will be zero for all
> >> the cores.
> >>
> >>>
> >>>> +        unsigned cores_per_node = DIV_ROUND_UP((x86ms->smp_dies *
> >>> smp_cores *
> >>>> +                                                smp_threads), nr_nodes);
> >>>
> >>> x86ms->smp_dies should be available at topo_info.dies_per_pkg,
> >>> smp_cores should available at topo_info.cores_per_die,
> >>> smp_threads should be available at topo_info.threads_per_core,
> >>> nr_nodes should be available at topo_info.nodes_per_pkg.
> >>>
> >>>>
> >>>>          /*
> >>>>           * die-id was optional in QEMU 4.0 and older, so keep it 
> >>>> optional
> >>>> @@ -1585,6 +1588,7 @@ static void pc_cpu_pre_plug(HotplugHandler
> >>> *hotplug_dev,
> >>>>          topo_ids.die_id = cpu->die_id;
> >>>>          topo_ids.core_id = cpu->core_id;
> >>>>          topo_ids.smt_id = cpu->thread_id;
> >>>> +        topo_ids.node_id = (cpu->core_id / cores_per_node) % nr_nodes;
> >>>
> >>> apicid_from_topo_ids() have access to topo_info and topo_ids, If
> >>> all the information you need to calculate node_id is already
> >>> available inside topo_info + topo_ids, we could be calculating it
> >>> inside apicid_from_topo_ids().  Why don't we do it?
> >>>
> >>> Also, is topo_ids.core_id really allowed to be larger than
> >>> cores_per_node when calling apicid_from_topo_ids()?
> >>
> >> Yes. It is. If we have two numa nodes and 8 cores. Then cores_per_node is
> >> 4. Nr_nodes =2. Yes. Core_id can be larger than cores_per_node.
> > 
> > I assumed core_id identified the core inside a die, and the range
> > would be [0, cores_per_die).  This seems to be what
> > apicid_from_topo_ids_epyc() expects.
> > 
> > We need clearer documentation on the semantics of each *_id
> > field, to avoid confusion.
> 
> Ok. Sure. I will add that. Can you please clarify on how to repro the
> issue. Marked the question at [2].

As it is usage of uninitialized memory, behavior is undefined and
not guaranteed to crash.  My QEMU binary was built with
--enable-debug, maybe it makes it easier to reproduce.

-- 
Eduardo