Re: [PATCH V2] irqchip/loongson-pch-pic: Update interrupt registration policy

[Thomas Gleixner]

On Sat, Mar 16 2024 at 15:03, Christophe JAILLET wrote:
> Le 16/03/2024 à 09:21, Tianyang Zhang a écrit :
>> @@ -171,6 +183,27 @@ static int pch_pic_domain_translate(struct irq_domain *d,
>>   			return -EINVAL;
>>   
>>   		*hwirq = fwspec->param[0] - priv->gsi_base;
>> +
>> +		raw_spin_lock_irqsave(&priv->pic_lock, flags);
>> +		/* Check pic-table to confirm if the hwirq has been assigned */
>> +		for (i = 0; i < priv->inuse; i++) {
>> +			if (priv->table[i] == *hwirq) {
>> +				*hwirq = i;
>> +				break;
>> +			}
>> +		}
>> +		if (i == priv->inuse) {
>> +			/* Assign a new hwirq in pic-table */
>> +			if (priv->inuse >= PIC_COUNT) {
>> +				pr_err("pch-pic domain has no free vectors\n");
>> +				raw_spin_unlock_irqrestore(&priv->pic_lock, flags);
>> +				return -EINVAL;
>> +			}
>> +			priv->table[priv->inuse] = *hwirq;
>> +			*hwirq = priv->inuse++;
>> +		}
>> +		raw_spin_unlock_irqrestore(&priv->pic_lock, flags);
>
> Hi,
>
> not sure if it helps or if this is a hot path, but, IIUC, taking the 
> lock could be avoided with some code reordering and 'inuse' being an 
> atomic_t.

It's the translate and setup path, so nothing to optimize here.