Re: [Qemu-devel] [PATCH v2 2/5] target/arm: Restrict pre-ARMv7 cpus to TCG

[Philippe Mathieu-Daudé]

On 9/3/19 2:10 PM, Thomas Huth wrote:
> On 03/09/2019 13.47, Philippe Mathieu-Daudé wrote:
>> KVM requires a cpu based on (at least) the ARMv7 architecture.
>>
>> The following CPUs are disabled:
>>
>> * ARMv4
>>
>>   - StrongARM (SA1100/1110)
>>   - OMAP1510 (TI925T)
>>
>> * ARMv5
>>
>>   - ARM926
>>   - ARM946
>>   - ARM1026
>>   - XScale (PXA250/255/260/261/262/270)
>>
>> * ARMv6
>>
>>   - ARM1136
>>   - ARM1176
>>   - ARM11MPCore
>>   - Cortex-M0
>>
>> Signed-off-by: Philippe Mathieu-Daudé <address@hidden>
>> ---
>> v2: fixed misplaced #endif (rth), list cpus
>> ---
>>  target/arm/cpu.c | 12 ++++++++++++
>>  1 file changed, 12 insertions(+)
>>
>> diff --git a/target/arm/cpu.c b/target/arm/cpu.c
>> index 2399c14471..f69780147c 100644
>> --- a/target/arm/cpu.c
>> +++ b/target/arm/cpu.c
>> @@ -1678,6 +1678,8 @@ static ObjectClass *arm_cpu_class_by_name(const char 
>> *cpu_model)
>>  /* CPU models. These are not needed for the AArch64 linux-user build. */
>>  #if !defined(CONFIG_USER_ONLY) || !defined(TARGET_AARCH64)
>>  
>> +#ifdef CONFIG_TCG
>> +
>>  static void arm926_initfn(Object *obj)
>>  {
>>      ARMCPU *cpu = ARM_CPU(obj);
>> @@ -1900,6 +1902,8 @@ static void cortex_m0_initfn(Object *obj)
>>      cpu->midr = 0x410cc200;
>>  }
>>  
>> +#endif
>> +
>>  static void cortex_m3_initfn(Object *obj)
>>  {
>>      ARMCPU *cpu = ARM_CPU(obj);
>> @@ -2283,6 +2287,8 @@ static void cortex_a15_initfn(Object *obj)
>>      define_arm_cp_regs(cpu, cortexa15_cp_reginfo);
>>  }
>>  
>> +#ifdef CONFIG_TCG
>> +
>>  static void ti925t_initfn(Object *obj)
>>  {
>>      ARMCPU *cpu = ARM_CPU(obj);
>> @@ -2451,6 +2457,8 @@ static void pxa270c5_initfn(Object *obj)
>>      cpu->reset_sctlr = 0x00000078;
>>  }
>>  
>> +#endif
>> +
>>  #ifndef TARGET_AARCH64
>>  /* -cpu max: if KVM is enabled, like -cpu host (best possible with this 
>> host);
>>   * otherwise, a CPU with as many features enabled as our emulation supports.
>> @@ -2523,6 +2531,7 @@ struct ARMCPUInfo {
>>  
>>  static const ARMCPUInfo arm_cpus[] = {
>>  #if !defined(CONFIG_USER_ONLY) || !defined(TARGET_AARCH64)
>> +#ifdef CONFIG_TCG
>>      { .name = "arm926",      .initfn = arm926_initfn },
>>      { .name = "arm946",      .initfn = arm946_initfn },
>>      { .name = "arm1026",     .initfn = arm1026_initfn },
>> @@ -2536,6 +2545,7 @@ static const ARMCPUInfo arm_cpus[] = {
>>      { .name = "arm11mpcore", .initfn = arm11mpcore_initfn },
>>      { .name = "cortex-m0",   .initfn = cortex_m0_initfn,
>>                               .class_init = arm_v7m_class_init },
>> +#endif
>>      { .name = "cortex-m3",   .initfn = cortex_m3_initfn,
>>                               .class_init = arm_v7m_class_init },
>>      { .name = "cortex-m4",   .initfn = cortex_m4_initfn,
>> @@ -2548,6 +2558,7 @@ static const ARMCPUInfo arm_cpus[] = {
>>      { .name = "cortex-a8",   .initfn = cortex_a8_initfn },
>>      { .name = "cortex-a9",   .initfn = cortex_a9_initfn },
>>      { .name = "cortex-a15",  .initfn = cortex_a15_initfn },
>> +#ifdef CONFIG_TCG
>>      { .name = "ti925t",      .initfn = ti925t_initfn },
>>      { .name = "sa1100",      .initfn = sa1100_initfn },
>>      { .name = "sa1110",      .initfn = sa1110_initfn },
>> @@ -2564,6 +2575,7 @@ static const ARMCPUInfo arm_cpus[] = {
>>      { .name = "pxa270-b1",   .initfn = pxa270b1_initfn },
>>      { .name = "pxa270-c0",   .initfn = pxa270c0_initfn },
>>      { .name = "pxa270-c5",   .initfn = pxa270c5_initfn },
>> +#endif
>>  #ifndef TARGET_AARCH64
>>      { .name = "max",         .initfn = arm_max_initfn },
>>  #endif
>>
> 
> All that #ifdeffery is a little bit ugly. I wonder whether we could
> solve this by moving the CPU registrations to separate files which then
> only get compiled if the corresponding CONFIG_ARM_Vx switch is set.

I tried splitting arm_cpus[] and move the ifdef in
arm_cpu_register_types(), but the resulting diff is much bigger and it
only reduces from 4 '#ifdef CONFIG_TCG' to 2...

I'll see what thinks Peter first.

> That reminds me of a patch series of mine where I tried to make the code
> compilable without CONFIG_ARM_V7M ... unfortunately, I never found
> enough spare time to finish and publish it... I'll have a try to see
> whether I can rebase it and send it as an RFC or so.

Well, this is the same goal of this series... So regarding on Peter's
comments I might wait on your work.

Regards,

Phil.