On Wed, 24 Jan 2024 23:31:58 +0100 Stefan Wahren wrote:there is actually another reason for this change which is not mentioned
The qca_spi driver create/stop the SPI kernel thread in caseWhat's the concern? I don't think that creating a thread is all
of netdev_open/close. This isn't optimal because there is no
need for such an expensive operation.
So improve this by moving create/stop of the SPI kernel into
the init/uninit ops. The open/close ops could just
'park/unpark' the SPI kernel thread.
expensive. And we shouldn't have a thread sitting around when
the interface isn't use. I mean - if you ask me what's better
a small chance that the creation will fail at open or having
a parked and unused thread when device is down - I'd pick
the former.. But I may well be missing the point.
Oops
@@ -825,6 +813,7 @@ static intI'm 90% sure this leaks resources on failure, too.
qcaspi_netdev_init(struct net_device *dev)
{
struct qcaspi *qca = netdev_priv(dev);
+ struct task_struct *thread;
dev->mtu = QCAFRM_MAX_MTU;
dev->type = ARPHRD_ETHER;
@@ -848,6 +837,15 @@ qcaspi_netdev_init(struct net_device *dev)
return -ENOBUFS;
}
+ thread = kthread_create(qcaspi_spi_thread, qca, "%s", dev->name);
+ if (IS_ERR(thread)) {
+ netdev_err(dev, "%s: unable to start kernel thread.\n",
+ QCASPI_DRV_NAME);
+ return PTR_ERR(thread);
Rest of the series LGTM!