| Message ID | 20170918160012.4317-4-david@redhat.com |
|---|---|
| State | New |
| Headers | show |
| Series | s390x: SMP for TCG (+ cleanups) | expand |
On Mon, 18 Sep 2017 17:59:48 +0200 David Hildenbrand <david@redhat.com> wrote: > core_id is not needed by linux-user, as the core_id a.k.a. CPU address > is only accessible from kernel space. > > Therefore, drop next_core_id and make cpu_index get autoassigned again > for linux-user. > > While at it, shield core_id and cpuid completely from linux-user. cpuid > can also only be queried from kernel space. > > Suggested-by: Igor Mammedov <imammedo@redhat.com> > Signed-off-by: David Hildenbrand <david@redhat.com> Reviewed-by: Igor Mammedov <imammedo@redhat.com> [...]
On 09/18/2017 08:59 AM, David Hildenbrand wrote: > /* sync cs->cpu_index and env->core_id. The latter is needed for TCG. */ > - cs->cpu_index = env->core_id; > + cs->cpu_index = cpu->env.core_id; > +#endif Any reason not to drop core_id entirely in favour of cpu_index? (Since cpu_index itself is generic and can't be dropped.) r~
On Mon, 25 Sep 2017 16:14:16 -0700 Richard Henderson <richard.henderson@linaro.org> wrote: > On 09/18/2017 08:59 AM, David Hildenbrand wrote: > > /* sync cs->cpu_index and env->core_id. The latter is needed for TCG. */ > > - cs->cpu_index = env->core_id; > > + cs->cpu_index = cpu->env.core_id; > > +#endif > > Any reason not to drop core_id entirely in favour of cpu_index? > (Since cpu_index itself is generic and can't be dropped.) I prefer core_id vs cpu_index as the later is too generic (and we were trying to decouple target specific attributes /coreid/cpuid/whatnot from cpu_index as much as possible so it would be more clear what they mean and if possible make them in-depended of QEMU internal cpu_index which is essentially not owned by target).
On 26.09.2017 10:49, Igor Mammedov wrote: > On Mon, 25 Sep 2017 16:14:16 -0700 > Richard Henderson <richard.henderson@linaro.org> wrote: > >> On 09/18/2017 08:59 AM, David Hildenbrand wrote: >>> /* sync cs->cpu_index and env->core_id. The latter is needed for TCG. */ >>> - cs->cpu_index = env->core_id; >>> + cs->cpu_index = cpu->env.core_id; >>> +#endif >> >> Any reason not to drop core_id entirely in favour of cpu_index? >> (Since cpu_index itself is generic and can't be dropped.) > I prefer core_id vs cpu_index as the later is too generic > (and we were trying to decouple target specific attributes > /coreid/cpuid/whatnot from cpu_index as much as possible > so it would be more clear what they mean and if possible > make them in-depended of QEMU internal cpu_index which is > essentially not owned by target). > As x86 also has this split I also prefer to keep it for now as it is. cpu_index just has to match our core id, otherwise e.g. migration could break (cpu_index is used as identifier for cpus in the migration stream). If we would drop core_id, we would have to implement a handler for the STAP instruction, as the CPU object is not accessible from translate.c. (not a big problem, just another reason to simply leave it as is for now), and also a new property handler, so only a uint16_t can be set using "core-id" onto a CPU object.
diff --git a/target/s390x/cpu-qom.h b/target/s390x/cpu-qom.h index 2e446fab51..0510c49e07 100644 --- a/target/s390x/cpu-qom.h +++ b/target/s390x/cpu-qom.h @@ -54,8 +54,6 @@ typedef struct S390CPUClass { bool is_migration_safe; const char *desc; - uint32_t next_core_id; - DeviceRealize parent_realize; void (*parent_reset)(CPUState *cpu); void (*load_normal)(CPUState *cpu); diff --git a/target/s390x/cpu.c b/target/s390x/cpu.c index 4920da3625..f42e28ea25 100644 --- a/target/s390x/cpu.c +++ b/target/s390x/cpu.c @@ -178,8 +178,9 @@ static void s390_cpu_realizefn(DeviceState *dev, Error **errp) { CPUState *cs = CPU(dev); S390CPUClass *scc = S390_CPU_GET_CLASS(dev); +#if !defined(CONFIG_USER_ONLY) S390CPU *cpu = S390_CPU(dev); - CPUS390XState *env = &cpu->env; +#endif Error *err = NULL; /* the model has to be realized before qemu_init_vcpu() due to kvm */ @@ -195,11 +196,6 @@ static void s390_cpu_realizefn(DeviceState *dev, Error **errp) max_cpus - 1); goto out; } -#else - /* implicitly set for linux-user only */ - cpu->env.core_id = scc->next_core_id; - scc->next_core_id++; -#endif if (cpu_exists(cpu->env.core_id)) { error_setg(&err, "Unable to add CPU with core-id: %" PRIu32 @@ -208,7 +204,9 @@ static void s390_cpu_realizefn(DeviceState *dev, Error **errp) } /* sync cs->cpu_index and env->core_id. The latter is needed for TCG. */ - cs->cpu_index = env->core_id; + cs->cpu_index = cpu->env.core_id; +#endif + cpu_exec_realizefn(cs, &err); if (err != NULL) { goto out; @@ -440,7 +438,9 @@ static gchar *s390_gdb_arch_name(CPUState *cs) } static Property s390x_cpu_properties[] = { +#if !defined(CONFIG_USER_ONLY) DEFINE_PROP_UINT32("core-id", S390CPU, env.core_id, 0), +#endif DEFINE_PROP_END_OF_LIST() }; diff --git a/target/s390x/cpu.h b/target/s390x/cpu.h index 9b549dc491..f13aea3dd8 100644 --- a/target/s390x/cpu.h +++ b/target/s390x/cpu.h @@ -150,8 +150,10 @@ struct CPUS390XState { CPU_COMMON +#if !defined(CONFIG_USER_ONLY) uint32_t core_id; /* PoP "CPU address", same as cpu_index */ uint64_t cpuid; +#endif uint64_t tod_offset; uint64_t tod_basetime; diff --git a/target/s390x/cpu_models.c b/target/s390x/cpu_models.c index 5169379db5..191aee6dd4 100644 --- a/target/s390x/cpu_models.c +++ b/target/s390x/cpu_models.c @@ -941,11 +941,13 @@ void s390_realize_cpu_model(CPUState *cs, Error **errp) apply_cpu_model(cpu->model, errp); +#if !defined(CONFIG_USER_ONLY) cpu->env.cpuid = s390_cpuid_from_cpu_model(cpu->model); if (tcg_enabled()) { /* basic mode, write the cpu address into the first 4 bit of the ID */ cpu->env.cpuid = deposit64(cpu->env.cpuid, 54, 4, cpu->env.core_id); } +#endif } static void get_feature(Object *obj, Visitor *v, const char *name,
core_id is not needed by linux-user, as the core_id a.k.a. CPU address is only accessible from kernel space. Therefore, drop next_core_id and make cpu_index get autoassigned again for linux-user. While at it, shield core_id and cpuid completely from linux-user. cpuid can also only be queried from kernel space. Suggested-by: Igor Mammedov <imammedo@redhat.com> Signed-off-by: David Hildenbrand <david@redhat.com> --- target/s390x/cpu-qom.h | 2 -- target/s390x/cpu.c | 14 +++++++------- target/s390x/cpu.h | 2 ++ target/s390x/cpu_models.c | 2 ++ 4 files changed, 11 insertions(+), 9 deletions(-)