Re: [PATCH v2 3/8] machine: Print supported CPU models instead of typenames

[Richard Henderson <richard.henderson@linaro.org>]

On 7/26/23 22:16, Gavin Shan wrote:
> 
> On 7/27/23 09:08, Richard Henderson wrote:
>> On 7/25/23 17:32, Gavin Shan wrote:
>>> -static const char *q800_machine_valid_cpu_types[] = {
>>> +static const char * const q800_machine_valid_cpu_types[] = {
>>>       M68K_CPU_TYPE_NAME("m68040"),
>>>       NULL
>>>   };
>>> +static const char * const q800_machine_valid_cpu_models[] = {
>>> +    "m68040",
>>> +    NULL
>>> +};
>>
>> I really don't like this replication.
>>
> 
> Right, it's going to be lots of replications, but gives much flexibility.
> There are 21 targets and we don't have fixed pattern for the mapping between
> CPU model name and CPU typename. I'm summarizing the used patterns like below.
> 
>    1 All CPU model names are mappinged to fixed CPU typename;
>    2 CPU model name is same to CPU typename;
>    3 CPU model name is alias to CPU typename;
>    4 CPU model name is prefix of CPU typename;
> 
>    Target         Categories    suffix-of-CPU-typename
>    -------------------------------------------------------
>    alpha          -234          -alpha-cpu
>    arm            ---4          -arm-cpu
>    avr            -2--
>    cris           --34          -cris-cpu
>    hexagon        ---4          -hexagon-cpu
>    hppa           1---
>    i386           ---4          -i386-cpu
>    loongarch      -2-4          -loongarch-cpu
>    m68k           ---4          -m68k-cpu
>    microblaze     1---
>    mips           ---4          -mips64-cpu  -mips-cpu
>    nios2          1---
>    openrisc       ---4          -or1k-cpu
>    ppc            --34          -powerpc64-cpu  -powerpc-cpu
>    riscv          ---4          -riscv-cpu
>    rx             -2-4          -rx-cpu
>    s390x          ---4          -s390x-cpu
>    sh4            --34          -superh-cpu
>    sparc          -2--
>    tricore        ---4          -tricore-cpu
>    xtensa         ---4          -xtensa-cpu

That is unfortunate, however...


> There are several options as below. Please let me know which one or something
> else is the best.
> 
> (a) Keep what we have and use mc->valid_{cpu_types, cpu_models}[] to track
> the valid CPU typenames and CPU model names.
> 
> (b) Introduce CPUClass::model_name_by_typename(). Every target has their own
> implementation to convert CPU typename to CPU model name. The CPU model name
> is parsed from mc->valid_cpu_types[i].
> 
>      char *CPUClass::model_by_typename(const char *typename);
> 
> (c) As we discussed before, use mc->valid_cpu_type_suffix and mc->valid_cpu_models
> because the CPU type check is currently needed by target arm/m68k/riscv where we
> do have fixed pattern to convert CPU model names to CPU typenames. The CPU typename
> is comprised of CPU model name and suffix. However, it won't be working when the CPU
> type check is required by other target where we have patterns other than this.

(d) Merge the two arrays together and use macro expansion, e.g.

typedef struct {
     const char *name;
     const char *type;
} Something;

#define ARM_SOMETHING(x)  { x, ARM_CPU_TYPE_NAME(x) }

static const Something valid[] = {
     ARM_SOMETHING("cortex-a53"),
     { NULL, NULL }
};

where Something ought to be better named.


r~