[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH 20/27] target-i386: do not call x86_cpu_realize(
From: |
Eduardo Habkost |
Subject: |
Re: [Qemu-devel] [PATCH 20/27] target-i386: do not call x86_cpu_realize() on cpu_x86_init() |
Date: |
Wed, 31 Oct 2012 15:01:12 -0200 |
User-agent: |
Mutt/1.5.21 (2010-09-15) |
On Wed, Oct 31, 2012 at 05:32:33PM +0100, Igor Mammedov wrote:
> On Wed, 24 Oct 2012 15:49:54 -0200
> Eduardo Habkost <address@hidden> wrote:
>
> > The PC code will need to run additional steps when initializing the CPU
> > object, before x86_cpu_realize(). So, make cpu_x86_init() not call
> Killing cpu_x86_init() altogether will make future re-factoring even easier.
> For present its code could be duplicated in cpu_init() and pc.c,
>
> and with cpu subclasses cpu_init () would be reduced to
> cpu = object_new(X86CPU.QEMUxx);
> cpu.realize();
I suspect *-user supports +feature,-feature on the CPU model string as
well. So both cases the cpu_model compat string parsing/property-setting
and the CPU class lookup would be necessary. So cpu_init() wouldn't look
so simple as above.
> and pc_cpus_init()
> cpu = object_new(X86CPU.QEMUxx);
> make cpu a child of /machine ();
> apply custom properties ();
> cpu.realize();
>
> I don't see any benefits in keeping cpu_x86_init() around and if we start
> touching it then just lets get rid of it in one step.
I believe the sequence that creates the CPU object will look like this
on *-user:
compat_normalize_cpu_model(cpu_model, &class_name, &features);
class = type_get_by_name(class_name);
cpu = object_new(class);
cpu_x86_set_props(cpu, features);
cpu.realize();
And on PC it will look like:
compat_normalize_cpu_model(cpu_model, &class_name, &features);
class = type_get_by_name(class_name);
cpu = object_new(class);
cpu_x86_set_props(cpu, features);
cpu_init_steps_sepcific_for_pc(cpu); /* APIC ID, make a child of
* /machine, whatever else.
*/
cpu.realize();
If cpu_init() was going to be just object_new() + cpu.realize(), I
wouldn't mind duplicating the code. But I don't see a reason to
duplicate code and not use a common function for the "cpu_model parsing
+ class lookup + object_new() + compat props setting" steps. I think
cpu_x86_init() can be that common function.
>
> > x86_cpu_realize(), and add two x86_cpu_realize() calls:
> >
> > - One on cpu_init(), that is called only by *-user
> > - One on pc_cpu_init(), that will include the more advanced PC CPU
> > initialization steps
> >
> > Signed-off-by: Eduardo Habkost <address@hidden>
> > ---
> > hw/pc.c | 12 +++++++++++-
> > target-i386/cpu.h | 14 ++++++++++++++
> > target-i386/helper.c | 11 ++++-------
> > 3 files changed, 29 insertions(+), 8 deletions(-)
> >
> > diff --git a/hw/pc.c b/hw/pc.c
> > index 85eab04..c209d3d 100644
> > --- a/hw/pc.c
> > +++ b/hw/pc.c
> > @@ -861,10 +861,20 @@ void pc_acpi_smi_interrupt(void *opaque, int irq, int
> > level)
> > static void pc_cpu_init(PCInitArgs *args, int cpu_index)
> > {
> > - if (!cpu_x86_init(args->qemu_args->cpu_model)) {
> > + Error *err = NULL;
> > + X86CPU *cpu;
> > +
> > + cpu = cpu_x86_init(args->qemu_args->cpu_model);
> > + if (!cpu) {
> > fprintf(stderr, "Unable to find x86 CPU definition\n");
> > exit(1);
> > }
> > +
> > + x86_cpu_realize(OBJECT(cpu), &err);
> > + if (err) {
> > + error_report("pc_cpu_init: %s\n", error_get_pretty(err));
> > + exit(1);
> > + }
> > }
> >
> > void pc_cpus_init(PCInitArgs *args)
> > diff --git a/target-i386/cpu.h b/target-i386/cpu.h
> > index 871c270..6853b17 100644
> > --- a/target-i386/cpu.h
> > +++ b/target-i386/cpu.h
> > @@ -21,6 +21,7 @@
> >
> > #include "config.h"
> > #include "qemu-common.h"
> > +#include "qemu-error.h"
> >
> > #ifdef TARGET_X86_64
> > #define TARGET_LONG_BITS 64
> > @@ -1008,12 +1009,25 @@ uint64_t cpu_get_tsc(CPUX86State *env);
> > #define TARGET_VIRT_ADDR_SPACE_BITS 32
> > #endif
> >
> > +/* Helper for simple CPU initialization (for target-independent code)
> > + *
> > + * Note that the PC code doesn't use this function, as it does additional
> > + * initialization steps between cpu_x86_init() and cpu_x86_realize() is
> > called.
> > + */
> > static inline CPUX86State *cpu_init(const char *cpu_model)
> > {
> > + Error *err = NULL;
> > X86CPU *cpu = cpu_x86_init(cpu_model);
> > if (cpu == NULL) {
> > return NULL;
> > }
> > +
> > + x86_cpu_realize(OBJECT(cpu), &err);
> > + if (err) {
> > + error_report("cpu_init: %s\n", error_get_pretty(err));
> > + return NULL;
> > + }
> > +
> > return &cpu->env;
> > }
> >
> > diff --git a/target-i386/helper.c b/target-i386/helper.c
> > index 1e5f61f..87a9221 100644
> > --- a/target-i386/helper.c
> > +++ b/target-i386/helper.c
> > @@ -1240,11 +1240,14 @@ int cpu_x86_get_descr_debug(CPUX86State *env,
> > unsigned int selector, return 1;
> > }
> >
> > +/* Initialize X86CPU object
> > + *
> > + * Callers must eventually call x86_cpu_realize(), to finish
> > initialization.
> > + */
> > X86CPU *cpu_x86_init(const char *cpu_model)
> > {
> > X86CPU *cpu;
> > CPUX86State *env;
> > - Error *err = NULL;
> >
> > cpu = X86_CPU(object_new(TYPE_X86_CPU));
> > env = &cpu->env;
> > @@ -1255,12 +1258,6 @@ X86CPU *cpu_x86_init(const char *cpu_model)
> > return NULL;
> > }
> >
> > - x86_cpu_realize(OBJECT(cpu), &err);
> > - if (err) {
> > - error_report("cpu_x86_init: %s\n", error_get_pretty(err));
> > - return NULL;
> > - }
> > -
> > return cpu;
> > }
> >
>
--
Eduardo
- [Qemu-devel] [PATCH 26/27] pc: create separate init function for pc-1.3, (continued)
- [Qemu-devel] [PATCH 26/27] pc: create separate init function for pc-1.3, Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 05/27] kvm: create kvm_arch_vcpu_id() function, Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 11/27] pc: add PC_DEFAULT_CPU_MODEL #define, Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 01/27] move I/O-related definitions from qemu-common.h to a new header (qemu-stdio.h), Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 20/27] target-i386: do not call x86_cpu_realize() on cpu_x86_init(), Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 23/27] pc: set fw_cfg data based on APIC ID calculation, Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 02/27] cpus.h: include qemu-stdio.h, Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 12/27] pc: add PCInitArgs parameter to pc_cpus_init(), Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 16/27] pc: pass PCInitArgs struct to pc_bios_init(), Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 21/27] fw_cfg: remove FW_CFG_MAX_CPUS from fw_cfg_init(), Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 25/27] target-i386: topology & APIC ID utility functions, Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 27/27] pc: generate APIC IDs according to CPU topology, Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 06/27] target-i386: kvm: set vcpu_id to APIC ID instead of CPU index, Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 22/27] pc: set CPU APIC ID explicitly, Eduardo Habkost, 2012/10/24
- [Qemu-devel] [PATCH 15/27] pc: rename bochs_bios_init() to pc_bios_init(), Eduardo Habkost, 2012/10/24