On 29/09/14 15:02, Konrad Rzeszutek Wilk wrote:
> On Sat, Sep 27, 2014 at 12:36:42AM +0800, Chen Gang wrote:
>> When xenbus_switch_state() fails, it will call xenbus_switch_fatal()
>
> Only on the first depth, not on the subsequent ones (as in if
> the first xenbus_switch_fail fails, it won't try to call
> xenbus_switch_state again and again).
>
>> internally, so need not return any status value, then use 'void' instead
>> of 'int' for xenbus_switch_state() and __xenbus_switch_state().
>
> When that switch occurs (to XenbusStateConnected) won't the watches
> fire - meaning we MUST make sure that the watch functions - if they
> use the xenbus_switch_state() they MUST not hold any locks - because
> they could be executed once more?
>
> Oh wait, we don't have to worry about that right now as the callbacks
> that pick up the messages from the XenBus are all gated on one mutex
> anyhow.
>
> Hm, anyhow, I would add this extra piece of information to the patch:
>
>
> diff --git a/drivers/xen/xen-pciback/xenbus.c b/drivers/xen/xen-pciback/xenbus.c
> index c214daa..f7399fd 100644
> --- a/drivers/xen/xen-pciback/xenbus.c
> +++ b/drivers/xen/xen-pciback/xenbus.c
> @@ -661,6 +661,12 @@ static void xen_pcibk_be_watch(struct xenbus_watch *watch,
>
> switch (xenbus_read_driver_state(pdev->xdev->nodename)) {
> case XenbusStateInitWait:
> + /*
> + * xenbus_switch_state can call xenbus_switch_fatal which will
> + * immediately set the state to XenbusStateClosing which
> + * means if we were reading for it here we MUST drop any
> + * locks so that we don't dead-lock.
> + */
Watches are asynchronous and serialised by the xenwatch thread. I can't
see what deadlock you're talking about here. Particularly since the
backend doesn't watch its own state node (it watches the frontend one).
> xen_pcibk_setup_backend(pdev);
> break;
>
>>
>> Also need be sure that all callers which check the return value must let
>> 'err' be 0.
>
> I am bit uncomfortable with that, that is due to:
>
>
> .. snip..
>> diff --git a/drivers/net/xen-netback/xenbus.c b/drivers/net/xen-netback/xenbus.c
>> index 9c47b89..b5c3d47 100644
>> --- a/drivers/net/xen-netback/xenbus.c
>> +++ b/drivers/net/xen-netback/xenbus.c
>> @@ -337,10 +337,7 @@ static int netback_probe(struct xenbus_device *dev,
>> if (err)
>> pr_debug("Error writing multi-queue-max-queues\n");
>>
>> - err = xenbus_switch_state(dev, XenbusStateInitWait);
>> - if (err)
>> - goto fail;
>> -
>> + xenbus_switch_state(dev, XenbusStateInitWait);
>
> Which if it fails it won't call:
>
> 354 fail:
> 355 pr_debug("failed\n");
> 356 netback_remove(dev);
> 357 return err;
>
>
> And since there is no watch on the backend state to go in Closing it won't
> ever call those and we leak memory.
It's not leaking the memory. All resources will be recovered when the
device is removed.
> The same is for xen-blkback mechanism in the probe function.
David
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
