Re: [kvm-unit-tests PATCH v2 3/3] lib: s390x: better smp interrupt checks

[Janosch Frank <frankja@linux.ibm.com>]

On 6/27/22 12:53, Claudio Imbrenda wrote:
> On Mon, 27 Jun 2022 11:28:18 +0200
> Janosch Frank <frankja@linux.ibm.com> wrote:
> 
>> On 6/24/22 16:45, Claudio Imbrenda wrote:
>>> Use per-CPU flags and callbacks for Program and Extern interrupts,
>>> instead of global variables.
>>>
>>> This allows for more accurate error handling; a CPU waiting for an
>>> interrupt will not have it "stolen" by a different CPU that was not
>>> supposed to wait for one, and now two CPUs can wait for interrupts at
>>> the same time.
>>>
>>> This will significantly improve error reporting and debugging when
>>> things go wrong.
>>>
>>> Both program interrupts and extern interrupts are now CPU-bound, even
>>> though some extern interrupts are floating (notably, the SCLP
>>> interrupt). In those cases, the testcases should mask interrupts and/or
>>> expect them appropriately according to need.
>>>
>>> Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
>>> ---
>>>    lib/s390x/asm/arch_def.h | 17 +++++++++++-
>>>    lib/s390x/smp.h          |  8 +-----
>>>    lib/s390x/interrupt.c    | 57 +++++++++++++++++++++++++++++-----------
>>>    lib/s390x/smp.c          | 11 ++++++++
>>>    4 files changed, 70 insertions(+), 23 deletions(-)
>> [...]
>>>    
>>> +struct lowcore *smp_get_lowcore(uint16_t idx)
>>> +{
>>> +	if (THIS_CPU->idx == idx)
>>> +		return &lowcore;
>>> +
>>> +	check_idx(idx);
>>> +	return cpus[idx].lowcore;
>>> +}
>>
>> This function is unused.
> 
> not currently, but it's useful to have in lib
> 
> should I split this into a separate patch?
> 
>>
>>> +
>>>    int smp_sigp(uint16_t idx, uint8_t order, unsigned long parm, uint32_t *status)
>>>    {
>>>    	check_idx(idx);
>>> @@ -253,6 +262,7 @@ static int smp_cpu_setup_nolock(uint16_t idx, struct psw psw)
>>>    
>>>    	/* Copy all exception psws. */
>>>    	memcpy(lc, cpus[0].lowcore, 512);
>>> +	lc->this_cpu = cpus + idx;
>>
>> Why not:
>> lc->this_cpu = &cpus[idx];
> 
> it's equivalent, do you have a reason for changing it?

It's more explicit.

> 
>>
>>>    
>>>    	/* Setup stack */
>>>    	cpus[idx].stack = (uint64_t *)alloc_pages(2);
>>> @@ -325,6 +335,7 @@ void smp_setup(void)
>>>    	for (i = 0; i < num; i++) {
>>>    		cpus[i].addr = entry[i].address;
>>>    		cpus[i].active = false;
>>> +		cpus[i].idx = i;
>>>    		/*
>>>    		 * Fill in the boot CPU. If the boot CPU is not at index 0,
>>>    		 * swap it with the one at index 0. This guarantees that the
>>
>