On Mon, 9 Mar 2020 14:04:39 -0500
Babu Moger <babu.mo...@amd.com> wrote:
> On 3/9/20 9:49 AM, Igor Mammedov wrote:
> > On Tue, 03 Mar 2020 13:58:03 -0600
> > Babu Moger <babu.mo...@amd.com> wrote:
> >
> >> Load the model specific handlers if available or else default handlers
> >> will be loaded. Add the model specific handlers if apicid decoding
> >> differs from the standard sequential numbering.
> >>
> >
> > this is still the old version of the patch and hadn't addressed feedback
> > from v4
>
> Yes. I was confused little bit about it. Will fix it.
>
> >
> >> Signed-off-by: Babu Moger <babu.mo...@amd.com>
> >> ---
> >> target/i386/cpu.c | 34 ++++++++++++++++++++++++++++++++++
> >> target/i386/cpu.h | 1 +
> >> 2 files changed, 35 insertions(+)
> >>
> >> diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> >> index c75cf744ab..f33d8b77f5 100644
> >> --- a/target/i386/cpu.c
> >> +++ b/target/i386/cpu.c
> >> @@ -51,6 +51,7 @@
> >> #include "sysemu/sysemu.h"
> >> #include "sysemu/tcg.h"
> >> #include "hw/qdev-properties.h"
> >> +#include "hw/i386/x86.h"
> > this dependency shouldn't be here, see below
>
> ok.
>
> >
> >> #include "hw/i386/topology.h"
> >> #ifndef CONFIG_USER_ONLY
> >> #include "exec/address-spaces.h"
> > [...]
> >> +void cpu_x86_init_apicid_fns(MachineState *machine)
> > it should be something like:
> > x86_use_epyc_apic_id_encoding(char *cpu_type)
> > try to avoid pulling in unnecessary dependency on Machine into cpu.c
>
> Ok.
>
> >
> >> +{
> >> + X86CPUClass *xcc =
> >> X86_CPU_CLASS(object_class_by_name(machine->cpu_type));
> >> + X86CPUModel *model = xcc->model;
> >> + X86CPUDefinition *def = model->cpudef;
> >> + X86MachineState *x86ms = X86_MACHINE(machine);
> >> +
> >> + if (def) {
> >> + if (def->apicid_from_cpu_idx) {
> >> + x86ms->apicid_from_cpu_idx = def->apicid_from_cpu_idx;
> >> + }
> >> + if (def->topo_ids_from_apicid) {
> >> + x86ms->topo_ids_from_apicid = def->topo_ids_from_apicid;
> >> + }
> >> + if (def->apicid_from_topo_ids) {
> >> + x86ms->apicid_from_topo_ids = def->apicid_from_topo_ids;
> >> + }
> >> + if (def->apicid_pkg_offset) {
> >> + x86ms->apicid_pkg_offset = def->apicid_pkg_offset;
> >> + }
> >> + }
> >> +}
> >
> > It was suggested to move defaults initialization to
> > x86_machine_class_init()
>
> ok. We don't need the above changes. I will use the boolean as you
> suggested and call this function in x86_cpus_init() to initialize the EPYC
> specific handler. Something similar like this below..
>
> x86_use_epyc_apic_id_encoding(char *cpu_type)
> {
> X86CPUClass *xcc = ... cpu_type ...
> return xcc->model->cpudef->use_epyc_apic_id_encoding
> }
>
> x86_cpus_init()
> {
> if (x86_use_epyc_apic_id_encoding(ms->cpu_type)) {
> x86ms->apicid_from_cpu_idx = ...epyc...
> x86ms->topo_ids_from_apicid = ...epyc...
> x86ms->apicid_from_topo_ids = ...epyc...
> x86ms->apicid_pkg_offset = ...epyc...
> }
> }
>
> Sounds right?
yes, something like this
>
> >
> > as was suggested at
> > [PATCH v4 12/16] hw/i386: Use the apicid handlers from X86MachineState
> > and acked by Eduardo
> >
> >> +
> >> static CPUCaches epyc_cache_info = {
> >> .l1d_cache = &(CPUCacheInfo) {
> >> .type = DATA_CACHE,
> >> diff --git a/target/i386/cpu.h b/target/i386/cpu.h
> >> index 20abbda647..34f0d994ef 100644
> >> --- a/target/i386/cpu.h
> >> +++ b/target/i386/cpu.h
> >> @@ -1895,6 +1895,7 @@ void cpu_clear_apic_feature(CPUX86State *env);
> >> void host_cpuid(uint32_t function, uint32_t count,
> >> uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t
> >> *edx);
> >> void host_vendor_fms(char *vendor, int *family, int *model, int
> >> *stepping);
> >> +void cpu_x86_init_apicid_fns(MachineState *machine);
> >>
> >> /* helper.c */
> >> bool x86_cpu_tlb_fill(CPUState *cs, vaddr address, int size,
> >>
> >
>
