Add a valid_cpu_types property
RFC 2/2 netduino2: Specify the valid CPUs
In reply to Alistair Francis---
hw/arm/netduino2.c | 5 +++++ 1 file changed, 5 insertions(+)
hw/arm/netduino2.c +5 -0
@@ -43,8 +43,13 @@ static void netduino2_init(MachineState *machine)static void netduino2_machine_init (MachineClass *mc) { +const char *val ="cortex-m3" ; + mc->desc ="Netduino 2 Machine" ; mc->init = netduino2_init; + + mc->valid_cpu_types =g_array_new (false, false,sizeof (char *)); +g_array_append_val (mc->valid_cpu_types, val); }DEFINE_MACHINE ("netduino2" , netduino2_machine_init)
Signature · 1 line
2.11.0
cover Add a valid_cpu_types property
There are numorous QEMU machines that only have a single or a handful of valid CPU options. To simplyfy the management of specificying which CPU is/isn't valid let's create a property that can be set in the machine init. We can then check to see if the user supplied CPU is in that list or not.
This is just a quick setup, if this method is agreed apon I can add a nice macro to add the valid CPU options (similar to SET_MACHINE_COMPAT) and improve the error messages. I just wanted to get some input before I spent too much time on that.
Alistair Francis (2):
machine: Add a valid_cpu_types property
netduino2: Specify the valid CPUs
hw/arm/netduino2.c | 5 +++++ hw/core/machine.c | 27 +++++++++++++++++++++++++++ include/hw/boards.h | 1 + 3 files changed, 33 insertions(+)
Signature · 1 line
2.11.0
RFC 1/2 machine: Add a valid_cpu_types property
---
hw/core/machine.c | 27 +++++++++++++++++++++++++++ include/hw/boards.h | 1 + 2 files changed, 28 insertions(+)
hw/core/machine.c +27 -0
@@ -758,6 +758,33 @@ void machine_run_board_init(MachineState *machine)machine_numa_finish_init (machine); } machine_class->init (machine); + +if (machine_class->valid_cpu_types && machine->cpu_model) { +const char *temp; +int i, len = machine_class->valid_cpu_types->len; + +for (i =0 ; i < len; i++) { + temp = g_array_index(machine_class->valid_cpu_types, char *, i); +if (!strcmp (machine->cpu_model, temp)) { +/* The user specificed CPU is in the valid field, we are +* good to go. +*/ +g_array_free (machine_class->valid_cpu_types, true); +return ; + } + } +/* The user specified CPU must not be a valid CPU, print a sane error */ + temp = g_array_index(machine_class->valid_cpu_types, char *, 0); +error_report ("Invalid CPU: %s" , machine->cpu_model); +error_printf ("The valid options are: %s" , temp); +for (i =1 ; i < len; i++) { + temp = g_array_index(machine_class->valid_cpu_types, char *, i); +error_printf (", %s" , temp); + } +error_printf ("\n" ); +g_array_free (machine_class->valid_cpu_types, true); +exit (1 ); + } }static void machine_class_finalize (ObjectClass *klass,void *data)
include/hw/boards.h +1 -0
@@ -172,6 +172,7 @@ struct MachineClass {int minimum_page_bits;bool has_hotpluggable_cpus;int numa_mem_align_shift; +GArray *valid_cpu_types;void (*numa_auto_assign_ram)(MachineClass *mc,NodeInfo *nodes,int nb_nodes,ram_addr_t size);
Signature · 1 line
2.11.0
Re: [Qemu-devel] [RFC v1 1/2] machine: Add a valid_cpu_types property
Quoted 12 lines · Alistair Francis
Signed-off-by Alistair Francis <[email protected]>---
hw/core/machine.c | 27 +++++++++++++++++++++++++++ include/hw/boards.h | 1 + 2 files changed, 28 insertions(+)hw/core/machine.c +0
-0
hw/core/machine.c void machine_run_board_init(MachineState *machine) · line 764
Quoted 7 lines
hw/core/machine.c +4
-0@@ -758,6 +758,33 @@ void machine_run_board_init(MachineState *machine)machine_numa_finish_init (machine); } machine_class->init (machine); + +if (machine_class->valid_cpu_types && machine->cpu_model) { +const char *temp; +int i, len = machine_class->valid_cpu_types->len;
I suggest doing this after Igor's series that replaces the cpu_model field (full -cpu string) with cpu_type (only the CPU type name).
hw/core/machine.c void machine_run_board_init(MachineState *machine) · line 773
Quoted 9 lines
hw/core/machine.c +9
-0+ +for (i =0 ; i < len; i++) { + temp = g_array_index(machine_class->valid_cpu_types, char *, i); +if (!strcmp (machine->cpu_model, temp)) { +/* The user specificed CPU is in the valid field, we are +* good to go. +*/ +g_array_free (machine_class->valid_cpu_types, true); +return ;
I suggest checking for:
object_class_dynamic_cast(object_class_get_name(machine->cpu_type), type)
instead. This way machines could just enumerate a common parent type to all supported CPU models.
hw/core/machine.c void machine_run_board_init(MachineState *machine) · line 784
Quoted 11 lines
hw/core/machine.c +11
-0+ } + } +/* The user specified CPU must not be a valid CPU, print a sane error */ + temp = g_array_index(machine_class->valid_cpu_types, char *, 0); +error_report ("Invalid CPU: %s" , machine->cpu_model); +error_printf ("The valid options are: %s" , temp); +for (i =1 ; i < len; i++) { + temp = g_array_index(machine_class->valid_cpu_types, char *, i); +error_printf (", %s" , temp); + } +error_printf ("\n" );
Now we have a completely new method to list the valid CPU types in addition to arch_query_cpu_definitions() and list_cpus() (which are already a bit messy and need to share more code).
I think this should share code with "-cpu help"/query-cpu-definitions instead. This means arch_query_cpu_definitions() will need a MachineClass* argument, if the user wants only the CPU types supported by a specific machine type, but I think it would be an interesting improvement to query-cpu-definitions.
Quoted 11 lines
hw/core/machine.c +3
-0+g_array_free (machine_class->valid_cpu_types, true); +exit (1 ); + } }static void machine_class_finalize (ObjectClass *klass,void *data)include/hw/boards.h +0
-0
include/hw/boards.h struct MachineClass { · line 175
Quoted 4 lines
include/hw/boards.h +1
-0@@ -172,6 +172,7 @@ struct MachineClass {int minimum_page_bits;bool has_hotpluggable_cpus;int numa_mem_align_shift; +GArray *valid_cpu_types;
The list of CPU types for a machine are very likely to be statically defined at build time. Any specific reason to not make it a simple char** pointer?
Quoted 6 lines
include/hw/boards.h +0
-0void (*numa_auto_assign_ram)(MachineClass *mc,NodeInfo *nodes,int nb_nodes,ram_addr_t size);Signature · 1 line
2.11.0
Signature · 1 line
Eduardo
Re: [Qemu-devel] [RFC v1 1/2] machine: Add a valid_cpu_types property
Quoted 8 lines · On Sat, 9 Sep 2017 17:16:57 -0300 Eduardo Habkost <[email protected]> wrote:
Quoted 7 lines · Alistair Francis
Signed-off-by Alistair Francis <[email protected]>---
hw/core/machine.c | 27 +++++++++++++++++++++++++++ include/hw/boards.h | 1 + 2 files changed, 28 insertions(+)
Quoted 16 lines
hw/core/machine.c void machine_run_board_init(MachineState *machine) · line 764
Quoted 12 lines
hw/core/machine.c +4
-0@@ -758,6 +758,33 @@ void machine_run_board_init(MachineState *machine)machine_numa_finish_init (machine); } machine_class->init (machine); + +if (machine_class->valid_cpu_types && machine->cpu_model) { +const char *temp; +int i, len = machine_class->valid_cpu_types->len;I suggest doing this after Igor's series that replaces the
cpu_modelfield (full -cpu string) withcpu_type(only the CPU type name).
in addition check should run before
machine_class->init(machine);
it should be possible after cpu_model -> cpu_type refactoring is complete
[...]
Quoted 5 lines
Quoted 5 lines
include/hw/boards.h +0
-0@@ -172,6 +172,7 @@ struct MachineClass {
Quoted 8 lines
include/hw/boards.h struct MachineClass { · line 175
Quoted 4 lines
include/hw/boards.h +1
-0int minimum_page_bits;bool has_hotpluggable_cpus;int numa_mem_align_shift; +GArray *valid_cpu_types;The list of CPU types for a machine are very likely to be statically defined at build time. Any specific reason to not make it a simple char** pointer?
wile working on cpu_model, I've seen only static checks used in some machines. So +1 to static null terminated array [
TYPE_FOO1,
TYPE_FOO2,
NULL
]
Quoted 8 lines
Quoted 7 lines
include/hw/boards.h +0
-0void (*numa_auto_assign_ram)(MachineClass *mc,NodeInfo *nodes,int nb_nodes,ram_addr_t size);Signature · 1 line
2.11.0
Re: [Qemu-devel] [RFC v1 2/2] netduino2: Specify the valid CPUs
In reply to Alistair FrancisQuoted 12 lines · Alistair Francis
Signed-off-by Alistair Francis <[email protected]>---
hw/arm/netduino2.c | 5 +++++ 1 file changed, 5 insertions(+)hw/arm/netduino2.c +0
-0@@ -43,8 +43,13 @@ static void netduino2_init(MachineState *machine)
hw/arm/netduino2.c static void netduino2_init(MachineState *machine) · line 53
Quoted 10 lines
hw/arm/netduino2.c +5
-0static void netduino2_machine_init (MachineClass *mc) { +const char *val ="cortex-m3" ; + mc->desc ="Netduino 2 Machine" ; mc->init = netduino2_init; + + mc->valid_cpu_types =g_array_new (false, false,sizeof (char *)); +g_array_append_val (mc->valid_cpu_types, val); }
Hi Alistair,
Nice idea! Why not adding an helper function which take a const char** in argument?
Fred
Quoted 3 lines
hw/arm/netduino2.c +0
-0DEFINE_MACHINE ("netduino2" , netduino2_machine_init)
Re: [Qemu-devel] [RFC v1 1/2] machine: Add a valid_cpu_types property
In reply to Eduardo HabkostQuoted 56 lines · Eduardo Habkost
Quoted 12 lines · Alistair Francis
Signed-off-by Alistair Francis <[email protected]>---
hw/core/machine.c | 27 +++++++++++++++++++++++++++ include/hw/boards.h | 1 + 2 files changed, 28 insertions(+)hw/core/machine.c +0
-0hw/core/machine.c void machine_run_board_init(MachineState *machine) · line 764
Quoted 7 lines
hw/core/machine.c +4
-0@@ -758,6 +758,33 @@ void machine_run_board_init(MachineState *machine)machine_numa_finish_init (machine); } machine_class->init (machine); + +if (machine_class->valid_cpu_types && machine->cpu_model) { +const char *temp; +int i, len = machine_class->valid_cpu_types->len;I suggest doing this after Igor's series that replaces the
cpu_modelfield (full -cpu string) withcpu_type(only the CPU type name).hw/core/machine.c void machine_run_board_init(MachineState *machine) · line 773
Quoted 9 lines
hw/core/machine.c +9
-0+ +for (i =0 ; i < len; i++) { + temp = g_array_index(machine_class->valid_cpu_types, char *, i); +if (!strcmp (machine->cpu_model, temp)) { +/* The user specificed CPU is in the valid field, we are +* good to go. +*/ +g_array_free (machine_class->valid_cpu_types, true); +return ;I suggest checking for:
object_class_dynamic_cast(object_class_get_name(machine->cpu_type), type)
instead. This way machines could just enumerate a common parent type to all supported CPU models.hw/core/machine.c void machine_run_board_init(MachineState *machine) · line 784
Quoted 11 lines
hw/core/machine.c +11
-0+ } + } +/* The user specified CPU must not be a valid CPU, print a sane error */ + temp = g_array_index(machine_class->valid_cpu_types, char *, 0); +error_report ("Invalid CPU: %s" , machine->cpu_model); +error_printf ("The valid options are: %s" , temp); +for (i =1 ; i < len; i++) { + temp = g_array_index(machine_class->valid_cpu_types, char *, i); +error_printf (", %s" , temp); + } +error_printf ("\n" );Now we have a completely new method to list the valid CPU types in addition to
arch_query_cpu_definitions()andlist_cpus()(which are already a bit messy and need to share more code).
Quoted 6 lines
I think this should share code with "-cpu help"/query-cpu-definitions instead. This means
arch_query_cpu_definitions()will need a MachineClass* argument, if the user wants only the CPU types supported by a specific machine type, but I think it would be an interesting improvement to query-cpu-definitions.
I don't see the improvement here. arch_query_cpu_definitions() is just a pretty simple list of that arch's CPUs. I don't really see how it helps in this case. We don't re-use much code at all.
Ideally I would like to see less arch dependent code in QEMU, and relying on arch_query_cpu_definitions() is moving in the opposite direction. It looks like arch_query_cpu_definitions() isn't supported by every architecture as well. I only see ARM, x86, PPC and S390x support.
I have addressed all the other comments so I'm going to send an RFCv2 out later today. The code looks a lot nicer now. I'm happy to keep discussing this though, just want to keep the ball rolling.
Thanks,
Alistair
Quoted 30 lines
Quoted 12 lines
hw/core/machine.c +3
-0+g_array_free (machine_class->valid_cpu_types, true); +exit (1 ); + } }static void machine_class_finalize (ObjectClass *klass,void *data)include/hw/boards.h +0
-0include/hw/boards.h struct MachineClass { · line 175
Quoted 4 lines
include/hw/boards.h +1
-0@@ -172,6 +172,7 @@ struct MachineClass {int minimum_page_bits;bool has_hotpluggable_cpus;int numa_mem_align_shift; +GArray *valid_cpu_types;The list of CPU types for a machine are very likely to be statically defined at build time. Any specific reason to not make it a simple char** pointer?
Quoted 6 lines
include/hw/boards.h +0
-1void (*numa_auto_assign_ram)(MachineClass *mc,NodeInfo *nodes,int nb_nodes,ram_addr_t size); --2.11.0
--
Eduardo
Re: [Qemu-devel] [RFC v1 2/2] netduino2: Specify the valid CPUs
In reply to KONRAD FredericQuoted 29 lines · KONRAD Frederic
Quoted 12 lines · Alistair Francis
Signed-off-by Alistair Francis <[email protected]>---
hw/arm/netduino2.c | 5 +++++ 1 file changed, 5 insertions(+)hw/arm/netduino2.c +0
-0hw/arm/netduino2.c static void netduino2_init(MachineState *machine) · line 52
Quoted 10 lines
hw/arm/netduino2.c +5
-0@@ -43,8 +43,13 @@ static void netduino2_init(MachineState *machine)static void netduino2_machine_init (MachineClass *mc) { +const char *val ="cortex-m3" ; + mc->desc ="Netduino 2 Machine" ; mc->init = netduino2_init; + + mc->valid_cpu_types =g_array_new (false, false,sizeof (char *)); +g_array_append_val (mc->valid_cpu_types, val); }Hi Alistair,
Quoted 2 lines
Nice idea! Why not adding an helper function which take a const char** in argument?
Thanks Fred.
I'm going to send a v2 with just that :)
Thanks,
Alistair
Quoted 6 lines
Fred
Quoted 2 lines
hw/arm/netduino2.c +0
-0DEFINE_MACHINE ("netduino2" , netduino2_machine_init)