Re: [patch 13/13] ptp: Convert ptp_open/read() to __free()

[Thomas Gleixner <tglx@linutronix.de>]

On Tue, Jun 24 2025 at 11:48, Paolo Abeni wrote:
> On 6/20/25 3:24 PM, Thomas Gleixner wrote:
>>  	scoped_guard(spinlock_irq, &queue->lock) {
>> -		size_t qcnt = queue_cnt(queue);
>> -
>> -		if (cnt > qcnt)
>> -			cnt = qcnt;
>> +		size_t qcnt = min((size_t)queue_cnt(queue), cnt / sizeof(*event));
>>  
>> -		for (size_t i = 0; i < cnt; i++) {
>> +		for (size_t i = 0; i < qcnt; i++) {
>>  			event[i] = queue->buf[queue->head];
>>  			/* Paired with READ_ONCE() in queue_cnt() */
>>  			WRITE_ONCE(queue->head, (queue->head + 1) % PTP_MAX_TIMESTAMPS);
>>  		}
>>  	}
>>  
>> -	cnt = cnt * sizeof(struct ptp_extts_event);
>> -
>> -	result = cnt;
>> -	if (copy_to_user(buf, event, cnt)) {
>> -		result = -EFAULT;
>> -		goto free_event;
>> -	}
>> -
>> -free_event:
>> -	kfree(event);
>> -exit:
>> -	return result;
>> +	return copy_to_user(buf, event, cnt) ? -EFAULT : cnt;
>>  }
>
> I'm likely low on coffee, but it looks like the above code is now
> returning the original amount of provided events, while the existing
> code returns the value bounded to the number of queue events.

Duh. Indeed. Well spotted.

Thanks,

        tglx