Re: [PATCH 01/16] hw/core/cpu: Let CPU object have a clock source

[Philippe Mathieu-Daudé]

On 10/6/20 8:11 PM, Philippe Mathieu-Daudé wrote:
> On 10/5/20 9:22 PM, Eduardo Habkost wrote:
>> On Mon, Oct 05, 2020 at 08:29:24PM +0200, Philippe Mathieu-Daudé wrote:
>>> On 10/5/20 8:09 PM, Philippe Mathieu-Daudé wrote:
>>>> On 10/5/20 7:44 PM, Eduardo Habkost wrote:
>>>>> On Mon, Oct 05, 2020 at 06:40:09PM +0200, Igor Mammedov wrote:
>>>>>> On Wed, 30 Sep 2020 12:16:53 +0200
>>>>>> Philippe Mathieu-Daudé <f4bug@amsat.org> wrote:
>>>>>>
>>>>>>> +arm/ppc/riscv folks
>>>>>>>
>>>>>>> On 9/30/20 9:43 AM, Igor Mammedov wrote:
>>>>>>>> On Mon, 28 Sep 2020 19:15:24 +0200
>>>>>>>> Philippe Mathieu-Daudé <f4bug@amsat.org> wrote:
>>>>>>>>   
>>>>>>>>> Let CPUState have a clock source (named 'clk') and CPUClass
>>>>>
>>>>> The language here confuses me: is this a clock source inside the
>>>>> CPU, or just a clock input that can be connected to a clock
>>>>> source somewhere?
>>>>
>>>> 2nd description, "somewhere". I'll reword.
>>>>
>>>>>
>>>>> See also comment below[1].
>>>>>
>>>>>>>>> have a clock_update() callback. The clock can be optionally
>>>>>>>>> set Using qdev_connect_clock_in() from the Clock API.
>>>>>>>>> If the clock changes, the optional clock_update() will be
>>>>>>>>> called.  
>>>>>
>>>>> What does "clock change" means?  Is this just about the
>>>>> frequency, or something else?
>>>>
>>>> A frequency changes -- which can be because a parent (in the
>>>> clock tree) changed its source using a MUX.
>>>>
>>>>>
>>>>> (By reading the Clock API documentation, it looks like it only
>>>>> represents the clock frequency, but I'm not sure)
>>>>>
>>>>>>>>
>>>>>>>> the sole user of it is mips cpu, so question is why
>>>>>>>> you are making it part of generic CPUm instead of
>>>>>>>> MIPSCPUClass/MIPSCPU?  
>>>>>>>
>>>>>>> This is a feature of the CPU, regardless its architecture.
>>>>>>>
>>>>>>> I expect the other archs to start using it soon.
>>>>>>
>>>>>> if there aren't any plans to actually to do that,
>>>>>> I'd keep it to MIPS class and generalize later when there is demand.
>>>>
>>>> No problem.
>>>>
>>>>>
>>>>> I normally don't mind if a feature is generic from the beginning.
>>>>> But in this case I'm inclined to agree with Igor.  Unless we
>>>>> expect to see arch-independent code to use CPUState.clock soon
>>>>> (do we?), having CPUState.clock existing but unused by most
>>>>> architectures would be misleading.
>>>>>
>>>>> Also, at least on x86 there are so many different clock sources,
>>>>> that I'm not sure it would be a desirable to have a generic clock
>>>>> input named "clk".
>>>>
>>>> Well X86 is the arch I'm less confident with. Anyhow if it has
>>>> multiple clock sources, I'd expect a Clock MUX block to select
>>>> an unique clock to feed the CPU.
>>>>
>>>>>
>>>>>>  
>>>>>>>
>>>>>>>>   
>>>>>>>>>
>>>>>>>>> Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
>>>>>>>>> ---
>>>>>>>>>  include/hw/core/cpu.h |  5 +++++
>>>>>>>>>  hw/core/cpu.c         | 12 ++++++++++++
>>>>>>>>>  2 files changed, 17 insertions(+)
>>>>>>>>>
>>>>>>>>> diff --git a/include/hw/core/cpu.h b/include/hw/core/cpu.h
>>>>>>>>> index 6c34798c8b3..6989d90c193 100644
>>>>>>>>> --- a/include/hw/core/cpu.h
>>>>>>>>> +++ b/include/hw/core/cpu.h
>>>>>>>>> @@ -31,6 +31,7 @@
>>>>>>>>>  #include "qemu/thread.h"
>>>>>>>>>  #include "qemu/plugin.h"
>>>>>>>>>  #include "qom/object.h"
>>>>>>>>> +#include "hw/clock.h"
>>>>>>>>>  
>>>>>>>>>  typedef int (*WriteCoreDumpFunction)(const void *buf, size_t size,
>>>>>>>>>                                       void *opaque);
>>>>>>>>> @@ -155,6 +156,7 @@ struct TranslationBlock;
>>>>>>>>>   * @disas_set_info: Setup architecture specific components of 
>>>>>>>>> disassembly info
>>>>>>>>>   * @adjust_watchpoint_address: Perform a target-specific adjustment 
>>>>>>>>> to an
>>>>>>>>>   * address before attempting to match it against watchpoints.
>>>>>>>>> + * @clock_update: Callback for input clock changes
>>>>>>>>>   *
>>>>>>>>>   * Represents a CPU family or model.
>>>>>>>>>   */
>>>>>>>>> @@ -176,6 +178,7 @@ struct CPUClass {
>>>>>>>>>                                    unsigned size, MMUAccessType 
>>>>>>>>> access_type,
>>>>>>>>>                                    int mmu_idx, MemTxAttrs attrs,
>>>>>>>>>                                    MemTxResult response, uintptr_t 
>>>>>>>>> retaddr);
>>>>>>>>> +    void (*clock_update)(CPUState *cpu);
>>>>>>>>>      bool (*virtio_is_big_endian)(CPUState *cpu);
>>>>>>>>>      int (*memory_rw_debug)(CPUState *cpu, vaddr addr,
>>>>>>>>>                             uint8_t *buf, int len, bool is_write);
>>>>>>>>> @@ -316,6 +319,7 @@ struct qemu_work_item;
>>>>>>>>>   *   QOM parent.
>>>>>>>>>   * @nr_cores: Number of cores within this CPU package.
>>>>>>>>>   * @nr_threads: Number of threads within this CPU.
>>>>>>>>> + * @clock: this CPU source clock (an output clock of another device)
>>>>>
>>>>> [1]
>>>>>
>>>>> What does "source clock" means?  Is this the same as "clock input"?
>>>>
>>>> Yes, for clocks it is common to use source/sink instead of input/output.
>>>> I'll try to reword.
>>>
>>> Hard to reword when it looks clear to oneself...
>>>
>>> @clock is the source, @cpu is the sink.
>>> @clock clocks @cpu at some frequency.
>>>
>>> One output from @clock is the @cpu.
>>> The @cpu has an unique input: @clock.
>>
>> The interchangeable usage of "clock source" and "clock input" is
>> what confuses me here.  CPUState.clock seems to be a clock input,
>> which may or may not be connected to a clock source.
>>
>> You seem to imply that "clock source" and "clock input" are
>> synonymous, but that's not what I understand from the clock API
>> documentation.
> 
> Hmm the concepts are different.
> 
> From a clock generator point of view, the "output" is
> the point where the signal leaves the block, to be
> eventually consumed by a sink. The generator doesn't
> know what the sinks are made of:
> 
> +-----------+   +----------> Sink1
> |           |   |
> |           +---+ ClkOut1
> |           |   |
> |  ClkGen   |   +----------> Sink2
> |           |
> |           +--> ClkOut2
> |           |
> +-----------+
> 
> From a device point of view, the "input" is where
> the external signal (generated by a source) is
> connected. The device doesn't know what is the
> source made of:
> 
>                      +-----------+
>                      |           |
>                      |           |
> (Source)  ClkIn ----->  Device   |
>                      |           |
>                      |           |
>                      +-----------+
> 
> So input/output refer to interface to connect the
> clock signal.
> 
> Source/sink refer to the other object having a
> relation with this signal.

Well, we don't need source/sink complication for QEMU,
restricting to input/output should be sufficient and
simple enough.

> 
>>
>>>
>>> Damien/Peter/Luc, do you have better description suggestions?
>>>
>>>>
>>>>>
>>>>>
>>>>>>>>>   * @running: #true if CPU is currently running (lockless).
>>>>>>>>>   * @has_waiter: #true if a CPU is currently waiting for the 
>>>>>>>>> cpu_exec_end;
>>>>>>>>>   * valid under cpu_list_lock.
>>>>>>>>> @@ -400,6 +404,7 @@ struct CPUState {
>>>>>>>>>      int num_ases;
>>>>>>>>>      AddressSpace *as;
>>>>>>>>>      MemoryRegion *memory;
>>>>>>>>> +    Clock *clock;
>>>>>>>>>  
>>>>>>>>>      void *env_ptr; /* CPUArchState */
>>>>>>>>>      IcountDecr *icount_decr_ptr;
>>>>>>>>> diff --git a/hw/core/cpu.c b/hw/core/cpu.c
>>>>>>>>> index c55c09f734c..37fcff3ec64 100644
>>>>>>>>> --- a/hw/core/cpu.c
>>>>>>>>> +++ b/hw/core/cpu.c
>>>>>>>>> @@ -30,6 +30,7 @@
>>>>>>>>>  #include "qemu/qemu-print.h"
>>>>>>>>>  #include "sysemu/tcg.h"
>>>>>>>>>  #include "hw/boards.h"
>>>>>>>>> +#include "hw/qdev-clock.h"
>>>>>>>>>  #include "hw/qdev-properties.h"
>>>>>>>>>  #include "trace/trace-root.h"
>>>>>>>>>  #include "qemu/plugin.h"
>>>>>>>>> @@ -247,6 +248,16 @@ void cpu_reset(CPUState *cpu)
>>>>>>>>>      trace_guest_cpu_reset(cpu);
>>>>>>>>>  }
>>>>>>>>>  
>>>>>>>>> +static void cpu_clk_update(void *opaque)
>>>>>>>>> +{
>>>>>>>>> +    CPUState *cpu = opaque;
>>>>>>>>> +    CPUClass *cc = CPU_GET_CLASS(cpu);
>>>>>>>>> +
>>>>>>>>> +    if (cc->clock_update) {
>>>>>>>>> +        cc->clock_update(cpu);
>>>>>>>>> +    }
>>>>>>>>> +}
>>>>>>>>> +
>>>>>>>>>  static void cpu_common_reset(DeviceState *dev)
>>>>>>>>>  {
>>>>>>>>>      CPUState *cpu = CPU(dev);
>>>>>>>>> @@ -367,6 +378,7 @@ static void cpu_common_initfn(Object *obj)
>>>>>>>>>      /* the default value is changed by qemu_init_vcpu() for softmmu 
>>>>>>>>> */
>>>>>>>>>      cpu->nr_cores = 1;
>>>>>>>>>      cpu->nr_threads = 1;
>>>>>>>>> +    cpu->clock = qdev_init_clock_in(DEVICE(obj), "clk", 
>>>>>>>>> cpu_clk_update, cpu);
>>>>>>>>>  
>>>>>>>>>      qemu_mutex_init(&cpu->work_mutex);
>>>>>>>>>      QSIMPLEQ_INIT(&cpu->work_list);  
>>>>>>>>   
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>