[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH v4 3/3] hw/i386: Remove node_id, nr_nodes and nodes_per_pkg f
|
From: |
Igor Mammedov |
|
Subject: |
Re: [PATCH v4 3/3] hw/i386: Remove node_id, nr_nodes and nodes_per_pkg from topology |
|
Date: |
Thu, 20 Aug 2020 14:57:57 +0200 |
On Wed, 19 Aug 2020 17:42:58 -0500
Babu Moger <babu.moger@amd.com> wrote:
> On 8/19/20 7:18 AM, Igor Mammedov wrote:
> > On Fri, 14 Aug 2020 16:39:40 -0500
> > Babu Moger <babu.moger@amd.com> wrote:
> >
> >> Remove node_id, nr_nodes and nodes_per_pkg from topology. Use
> >> die_id, nr_dies and dies_per_pkg which is already available.
> >> Removes the confusion over two variables.
> >>
> >> With node_id removed in topology the uninitialized memory issue
> >> with -device and CPU hotplug will be fixed.
> >>
> >> Link:
> >> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fbugzilla.redhat.com%2Fshow_bug.cgi%3Fid%3D1828750&data=02%7C01%7Cbabu.moger%40amd.com%7C466254703c904bd4c86c08d8443a0e91%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637334363427680286&sdata=%2Fr7Mucca8Pr%2BrjvwJ6S9zxiIg3HCKAiPp4PvGP3nvZI%3D&reserved=0
> >> Signed-off-by: Babu Moger <babu.moger@amd.com>
> >> ---
> >> hw/i386/pc.c | 1 -
> >> hw/i386/x86.c | 1 -
> >> include/hw/i386/topology.h | 40 +++++++++-------------------------------
> >> target/i386/cpu.c | 11 +++--------
> >> target/i386/cpu.h | 1 -
> >> 5 files changed, 12 insertions(+), 42 deletions(-)
> >>
> >> diff --git a/hw/i386/pc.c b/hw/i386/pc.c
> >> index 47c5ca3e34..0ae5cb3af4 100644
> >> --- a/hw/i386/pc.c
> >> +++ b/hw/i386/pc.c
> >> @@ -1498,7 +1498,6 @@ static void pc_cpu_pre_plug(HotplugHandler
> >> *hotplug_dev,
> >> init_topo_info(&topo_info, x86ms);
> >>
> >> env->nr_dies = x86ms->smp_dies;
> >> - env->nr_nodes = topo_info.nodes_per_pkg;
> >> env->pkg_offset = x86ms->apicid_pkg_offset(&topo_info);
> >>
> >> /*
> >> diff --git a/hw/i386/x86.c b/hw/i386/x86.c
> >> index e90c42d2fc..4efa1f8b87 100644
> >> --- a/hw/i386/x86.c
> >> +++ b/hw/i386/x86.c
> >> @@ -62,7 +62,6 @@ inline void init_topo_info(X86CPUTopoInfo *topo_info,
> >> {
> >> MachineState *ms = MACHINE(x86ms);
> >>
> >> - topo_info->nodes_per_pkg = ms->numa_state->num_nodes /
> >> ms->smp.sockets;
> >> topo_info->dies_per_pkg = x86ms->smp_dies;
> >> topo_info->cores_per_die = ms->smp.cores;
> >> topo_info->threads_per_core = ms->smp.threads;
> >> diff --git a/include/hw/i386/topology.h b/include/hw/i386/topology.h
> >> index 07239f95f4..05ddde7ba0 100644
> >> --- a/include/hw/i386/topology.h
> >> +++ b/include/hw/i386/topology.h
> >> @@ -47,14 +47,12 @@ typedef uint32_t apic_id_t;
> >>
> >> typedef struct X86CPUTopoIDs {
> >> unsigned pkg_id;
> >> - unsigned node_id;
> >> unsigned die_id;
> >> unsigned core_id;
> >> unsigned smt_id;
> >> } X86CPUTopoIDs;
> >>
> >> typedef struct X86CPUTopoInfo {
> >> - unsigned nodes_per_pkg;
> >> unsigned dies_per_pkg;
> >> unsigned cores_per_die;
> >> unsigned threads_per_core;
> >> @@ -89,11 +87,6 @@ static inline unsigned apicid_die_width(X86CPUTopoInfo
> >> *topo_info)
> >> return apicid_bitwidth_for_count(topo_info->dies_per_pkg);
> >> }
> >>
> >> -/* Bit width of the node_id field per socket */
> >> -static inline unsigned apicid_node_width_epyc(X86CPUTopoInfo *topo_info)
> >> -{
> >> - return apicid_bitwidth_for_count(MAX(topo_info->nodes_per_pkg, 1));
> >> -}
> >> /* Bit offset of the Core_ID field
> >> */
> >> static inline unsigned apicid_core_offset(X86CPUTopoInfo *topo_info)
> >> @@ -114,30 +107,23 @@ static inline unsigned
> >> apicid_pkg_offset(X86CPUTopoInfo *topo_info)
> >> return apicid_die_offset(topo_info) + apicid_die_width(topo_info);
> >> }
> >>
> >> -#define NODE_ID_OFFSET 3 /* Minimum node_id offset if numa configured */
> >> +#define EPYC_DIE_OFFSET 3 /* Minimum die_id offset if numa configured */
> > ^^
> >
> > from EPYC's pov NUMA is always configured, there no 'if'
> >
> > but I have a question, is it possible to drop EPYC_DIE_OFFSET
> > and reuse apicid_die_offset(), will it break something?
>
> Yes. I am also thinking about it. We can go back to existing
> apicid_die_width().
Given we are using die_id now, we probably can get rid of all
topo hooks to generate apicid.
> Looking back again, I think all these code changes related to node_id is
> causing more issues than fixes. We have added all these code to handle
> some non- SPECed configurations like
> https://bugzilla.redhat.com/show_bug.cgi?id=1728166.
>
> Going forward it might create even more issues. Now, I think we should go
> back and remove all these changes and just use the default decoding. Most
> of the SPECed configuration would work just fine. With some non-SPECed
> user inputs, it will create some sub-optimal configuration. If we can live
> with that we will be just fine. I did not anticipate all these problems
> when I originally started this work. Sorry, my bad.
Topology code is complex and sometimes it's easier to add a new thing,
it's just that not every time it turns out as expected.
We overlooked posiblilty to reuse die-id, which has lead to more
complex and fragile outcome.
But it's fine, as long as it gets fixed in the end.
> > The reason for question is that, we (deviating from spec)
> > do allow for unbound core/threads number so die_id, already could
> > be shifted beyound ApicId[5:4], likewise we do allow for unbound
> > sockets so ApicId[6] is also out of spec.
> > If we are fine with ApicId being encoded in these cases out of
> > spec, then why should we insist on DIE_OFFSET minumum at all?
> > Why not just allow existing apicid_die_width() to encode die_id?
> >
> > In this case it will produce SPECed ApicId if user has provided
> > -smp that matches currently released EPYC's and in all other cases
> > it will be out of spec ApicId like when we set sockets/cores/trheads
> > to out of spec values.
> >
> >> /*
> >> - * Bit offset of the node_id field
> >> - *
> >> - * Make sure nodes_per_pkg > 0 if numa configured else zero.
> >> + * Bit offset of the die_id field
> >> */
> >> -static inline unsigned apicid_node_offset_epyc(X86CPUTopoInfo *topo_info)
> >> +static inline unsigned apicid_die_offset_epyc(X86CPUTopoInfo *topo_info)
> >> {
> >> - unsigned offset = apicid_die_offset(topo_info) +
> >> - apicid_die_width(topo_info);
> >> + unsigned offset = apicid_core_offset(topo_info) +
> >> + apicid_core_width(topo_info);
> >>
> >> - if (topo_info->nodes_per_pkg) {
> >> - return MAX(NODE_ID_OFFSET, offset);
> >> - } else {
> >> - return offset;
> >> - }
> >> + return MAX(EPYC_DIE_OFFSET, offset);
> >> }
> >>
> >> /* Bit offset of the Pkg_ID (socket ID) field */
> >> static inline unsigned apicid_pkg_offset_epyc(X86CPUTopoInfo *topo_info)
> >> {
> >> - return apicid_node_offset_epyc(topo_info) +
> >> - apicid_node_width_epyc(topo_info);
> >> + return apicid_die_offset_epyc(topo_info) +
> >> apicid_die_width(topo_info);
> >> }
> >>
> >> /*
> >> @@ -150,8 +136,7 @@ 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)) |
> >> - (topo_ids->die_id << apicid_die_offset(topo_info)) |
> >> + (topo_ids->die_id << apicid_die_offset_epyc(topo_info)) |
> >> (topo_ids->core_id << apicid_core_offset(topo_info)) |
> >> topo_ids->smt_id;
> >> }
> >> @@ -160,15 +145,11 @@ static inline void
> >> x86_topo_ids_from_idx_epyc(X86CPUTopoInfo *topo_info,
> >> unsigned cpu_index,
> >> X86CPUTopoIDs *topo_ids)
> >> {
> >> - unsigned nr_nodes = MAX(topo_info->nodes_per_pkg, 1);
> >> unsigned nr_dies = topo_info->dies_per_pkg;
> >> unsigned nr_cores = topo_info->cores_per_die;
> >> unsigned nr_threads = topo_info->threads_per_core;
> >> - unsigned cores_per_node = DIV_ROUND_UP((nr_dies * nr_cores *
> >> nr_threads),
> >> - nr_nodes);
> >>
> >> topo_ids->pkg_id = cpu_index / (nr_dies * nr_cores * nr_threads);
> >> - topo_ids->node_id = (cpu_index / cores_per_node) % nr_nodes;
> >> topo_ids->die_id = cpu_index / (nr_cores * nr_threads) % nr_dies;
> >> topo_ids->core_id = cpu_index / nr_threads % nr_cores;
> >> topo_ids->smt_id = cpu_index % nr_threads;
> >> @@ -188,11 +169,8 @@ static inline void
> >> x86_topo_ids_from_apicid_epyc(apic_id_t apicid,
> >> (apicid >> apicid_core_offset(topo_info)) &
> >> ~(0xFFFFFFFFUL << apicid_core_width(topo_info));
> >> topo_ids->die_id =
> >> - (apicid >> apicid_die_offset(topo_info)) &
> >> + (apicid >> apicid_die_offset_epyc(topo_info)) &
> >> ~(0xFFFFFFFFUL << apicid_die_width(topo_info));
> >> - topo_ids->node_id =
> >> - (apicid >> apicid_node_offset_epyc(topo_info)) &
> >> - ~(0xFFFFFFFFUL << apicid_node_width_epyc(topo_info));
> >> topo_ids->pkg_id = apicid >> apicid_pkg_offset_epyc(topo_info);
> >> }
> >>
> >> diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> >> index c892432cae..ba0a24f6b8 100644
> >> --- a/target/i386/cpu.c
> >> +++ b/target/i386/cpu.c
> >> @@ -345,7 +345,6 @@ static void encode_cache_cpuid8000001d(CPUCacheInfo
> >> *cache,
> >> uint32_t *ecx, uint32_t *edx)
> >> {
> >> uint32_t l3_cores;
> >> - unsigned nodes = MAX(topo_info->nodes_per_pkg, 1);
> >>
> >> assert(cache->size == cache->line_size * cache->associativity *
> >> cache->partitions * cache->sets);
> >> @@ -355,10 +354,9 @@ static void encode_cache_cpuid8000001d(CPUCacheInfo
> >> *cache,
> >>
> >> /* L3 is shared among multiple cores */
> >> if (cache->level == 3) {
> >> - l3_cores = DIV_ROUND_UP((topo_info->dies_per_pkg *
> >> - topo_info->cores_per_die *
> >> + l3_cores = DIV_ROUND_UP((topo_info->cores_per_die *
> >> topo_info->threads_per_core),
> >> - nodes);
> >> + topo_info->dies_per_pkg);
> >> *eax |= (l3_cores - 1) << 14;
> >> } else {
> >> *eax |= ((topo_info->threads_per_core - 1) << 14);
> >> @@ -387,7 +385,6 @@ static void encode_topo_cpuid8000001e(X86CPUTopoInfo
> >> *topo_info, X86CPU *cpu,
> >> uint32_t *ecx, uint32_t *edx)
> >> {
> >> X86CPUTopoIDs topo_ids = {0};
> >> - unsigned long nodes = MAX(topo_info->nodes_per_pkg, 1);
> >>
> >> x86_topo_ids_from_apicid_epyc(cpu->apic_id, topo_info, &topo_ids);
> >>
> >> @@ -433,7 +430,7 @@ static void encode_topo_cpuid8000001e(X86CPUTopoInfo
> >> *topo_info, X86CPU *cpu,
> >> * NodeId is combination of node and socket_id which is already
> >> decoded
> >> * in apic_id. Just use it by shifting.
> >> */
> >> - *ecx = ((nodes - 1) << 8) | cpu->node_id;
> >> + *ecx = ((topo_info->dies_per_pkg - 1) << 8) | cpu->node_id;
> >> *edx = 0;
> >> }
> >>
> >> @@ -5484,7 +5481,6 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index,
> >> uint32_t count,
> >> uint32_t signature[3];
> >> X86CPUTopoInfo topo_info;
> >>
> >> - topo_info.nodes_per_pkg = env->nr_nodes;
> >> topo_info.dies_per_pkg = env->nr_dies;
> >> topo_info.cores_per_die = cs->nr_cores;
> >> topo_info.threads_per_core = cs->nr_threads;
> >> @@ -6944,7 +6940,6 @@ static void x86_cpu_initfn(Object *obj)
> >> FeatureWord w;
> >>
> >> env->nr_dies = 1;
> >> - env->nr_nodes = 1;
> >> cpu_set_cpustate_pointers(cpu);
> >>
> >> object_property_add(obj, "family", "int",
> >> diff --git a/target/i386/cpu.h b/target/i386/cpu.h
> >> index e1a5c174dc..4c89bee8d1 100644
> >> --- a/target/i386/cpu.h
> >> +++ b/target/i386/cpu.h
> >> @@ -1629,7 +1629,6 @@ typedef struct CPUX86State {
> >> TPRAccess tpr_access_type;
> >>
> >> unsigned nr_dies;
> >> - unsigned nr_nodes;
> >> unsigned pkg_offset;
> >> } CPUX86State;
> >>
> >>
> >>
> >
>
[PATCH v4 1/3] i386: Simplify CPUID_8000_001E for AMD, Babu Moger, 2020/08/15
Re: [PATCH v4 0/3] Modify AMD topology to use socket/dies/core/thread model, no-reply, 2020/08/15